What Is Nominal vs. Ordinal Data?
Nominal and ordinal are two of the four levels of measurement in statistics, and they represent different ways of categorizing data. Nominal data classifies observations into unordered categories, like eye color, product type, or geographic region, where no category is inherently "higher" or "lower" than another. Ordinal data also uses categories, but those categories have a meaningful rank order, like education level, satisfaction rating, or income bracket. The critical difference is that ordinal data tells you the sequence, while nominal data only tells you the group. Understanding which type you're working with determines which statistical analyses are valid.
Why the Distinction Matters in Research
Using the wrong analysis for your data type produces misleading results. Calculating a mean for nominal data (the "average" gender or zip code) is meaningless, and treating ordinal data as though intervals are equal can distort findings. A 2019 study in BMC Medical Research Methodology found that approximately 46% of published health studies used inappropriate statistical tests for their measurement level, most commonly by treating ordinal data as interval.
How Nominal and Ordinal Data Work
Nominal Data
Nominal data is purely categorical. Each value represents a label with no inherent order or quantity. The word "nominal" comes from the Latin nomen (name), you're naming things, not ranking them.
Examples:
- Gender (male, female, non-binary)
- Blood type (A, B, AB, O)
- Survey response for "How did you hear about us?" (Google search, friend referral, social media, podcast)
- Product SKU or model name
You can count how many observations fall into each category and calculate proportions, but you can't rank the categories or calculate meaningful averages. The only valid measure of central tendency for nominal data is the mode (the most frequent category).
A special case is dichotomous (binary) nominal data, where there are only two categories: yes/no, male/female, purchased/didn't purchase.
Ordinal Data
Ordinal data has categories with a clear rank order, but the distances between categories aren't necessarily equal. You know that "Strongly Agree" is higher than "Agree," but you can't say the gap between them is the same as the gap between "Neutral" and "Disagree."
Examples:
- Education level (high school, bachelor's, master's, doctorate)
- Customer satisfaction (very dissatisfied, dissatisfied, neutral, satisfied, very satisfied)
- Competition placement (1st, 2nd, 3rd)
- T-shirt size (S, M, L, XL)
You can rank ordinal categories and identify the median (middle value), but calculating a mean is technically inappropriate because the intervals aren't defined. In practice, many researchers treat ordinal scales with five or more points as approximately interval, a pragmatic choice that most statisticians accept for large samples.
Side-by-Side Comparison
| Feature | Nominal | Ordinal |
|---|---|---|
| Categories | Yes | Yes |
| Rank order | No | Yes |
| Equal intervals | No | No |
| Valid central tendency | Mode | Mode, Median |
| Valid dispersion measure | Frequency distribution | Range, percentiles |
| Statistical tests | Chi-square, Fisher's exact | Mann-Whitney U, Kruskal-Wallis, Spearman's rho |
| Can calculate mean | No | Debated (common in practice) |
| Example | Product category | Satisfaction rating |
How to Identify Your Data Type
Ask yourself two questions:
- Can I rank the categories in a meaningful order? If no, it's nominal. If yes, move to question two.
- Are the gaps between categories equal and measurable? If no, it's ordinal. If yes, it's interval or ratio.
Sometimes the same variable can be measured at different levels depending on how you ask the question. Income measured as "under $50K / $50K-$100K / over $100K" is ordinal. Income measured as an exact dollar amount is ratio.
When to Use Each Type
- Use nominal scales when you need to segment respondents into groups for cross-tabulation (age bracket, region, customer type)
- Use ordinal scales when you need to measure attitudes, preferences, or subjective experiences that have a natural order
- Use nominal coding for open-ended response categories that don't have an inherent ranking
- Use ordinal scales when you want respondents to rank items from most to least preferred
- Choose interval or ratio scales when you need to calculate precise means and do parametric statistical testing
Common Mistakes to Avoid
- Calculating averages on nominal data: the mean of zip codes or product categories is nonsensical
- Assuming equal intervals in ordinal data: the distance between "satisfied" and "very satisfied" isn't necessarily the same as between "neutral" and "satisfied"
- Using Pearson correlation on ordinal data: use Spearman's rank correlation instead
- Collapsing ordinal categories arbitrarily: combining "strongly agree" and "agree" into a single "top-two box" score discards information about response intensity
How Quali-Fi Supports Both Data Types
Quali-Fi's survey platform automatically applies the correct analysis for each question type. Nominal questions (multiple choice, demographic selectors) get frequency tables and chi-square testing in cross-tabs, while ordinal questions (Likert scales, ranking items) display median scores, distribution charts, and non-parametric significance tests. The platform supports all four measurement levels across 50+ question types, starting at $89/month.
Frequently Asked Questions
Is a Likert scale nominal or ordinal?
A Likert scale is ordinal. The response options (e.g., "Strongly Disagree" to "Strongly Agree") have a clear rank order, but the distances between adjacent options aren't guaranteed to be equal. That said, most researchers treat Likert scales with five or more points as approximately interval for practical analysis.
Can ordinal data be converted to nominal?
Yes, but you lose information. You can collapse an ordinal satisfaction scale into nominal groups like "satisfied" and "not satisfied," but you discard the gradation between levels. Go in this direction only when you have a specific analytical reason, like creating binary segments for logistic regression.
What level of measurement is a yes/no question?
A yes/no question produces nominal (specifically dichotomous) data. There's no inherent order, "yes" isn't higher than "no" in a universal sense, though in specific contexts (like purchase intent), you might assign numerical values for analysis.